In this page, we demonstrate the output of various oversampling and noise removal techniques, using default parameters.
For binary oversampling and nosie removal, an artificial database was used, available in the datasets module.
In the captions of the images some abbreviations referring to the operating principles are placed. Namely:
- NR: noise removal is involved
- DR: dimension reduction is applied
- Clas: some supervised classifier is used
- SCmp: sampling is carried out componentwise (attributewise)
- SCpy: sampling is carried out by copying instances
- SO: ordinary sampling (just like in SMOTE)
- M: memetic optimization is used
- DE: density estimation is used
- DB: density based - the sampling is based on a density of importance assigned to the instances
- Ex: the sampling is extensive - samples are added successively, not optimizing the holistic distribution of a given number of samples
- CM: changes majority - even majority samples can change
- Clus: uses some clustering technique
- BL: identifies and samples the neighborhoods of borderline samples
- A: developed for a specific application
